Structure-Based Optimization of Potent and Selective Inhibitors of the

Structure-Based Optimization of Potent and Selective Inhibitors of the Tyrosine Kinase Erythropoietin Producing Human Hepatocellular Carcinoma Recepto...
0 downloads 4 Views 1MB Size
J. Med. Chem. 2009, 52, 6433–6446 6433 DOI: 10.1021/jm9009444

Structure-Based Optimization of Potent and Selective Inhibitors of the Tyrosine Kinase Erythropoietin Producing Human Hepatocellular Carcinoma Receptor B4 (EphB4) Karine Lafleur,†,‡ Danzhi Huang,*,† Ting Zhou,† Amedeo Caflisch,*,† and Cristina Nevado*,‡ †

Department of Biochemistry and ‡Department of Organic Chemistry, University of Zurich, Winterthurerstrasse 190, CH-8057, Zurich, Switzerland Received June 26, 2009

The tyrosine kinase EphB4 is an attractive target for drug design because of its recognized role in cancerrelated angiogenesis. Recently, a series of commercially available xanthine derivatives were identified as micromolar inhibitors of EphB4 by high-throughput fragment-based docking into the ATP-binding site of the kinase domain. Here, we have exploited the binding mode obtained by automatic docking for the optimization of these EphB4 inhibitors by chemical synthesis. Addition of only two heavy atoms, methyl and hydroxyl groups, to compound 4 has yielded the single-digit nanomolar inhibitor 66, with a remarkable improvement of the ligand efficiency from 0.26 to 0.37 kcal/(mol per non-hydrogen atom). Compound 66 shows very high affinity for a few other tyrosine kinases with threonine as gatekeeper residue (Abl, Lck, and Src). On the other hand, it is selective against kinases with a larger gatekeeper. A 45 ns molecular dynamics (MD) simulation of the complex of EphB4 and compound 66 provides further validation of the binding mode obtained by fragment-based docking.

*To whom correspondence should be addressed. For D.H.: phone, (41) 446355568; fax, (41) 446356862; e-mail, [email protected]. For A.C.: phone, (41) 446355521; fax, (41) 446356862; e-mail, caflisch@ bioc.uzh.ch. For C.N.: phone, (41) 446353945; fax, (41) 446353948; e-mail, [email protected]. a Abbreviations: Abl, Abelson murine leukemia viral oncogene homologue; ALTA, anchor-based library tailoring; ATP, adenosine triphosphate; DIPEA, diisopropylethylamine; DMF, dimethylformamide; DMSO, dimethyl sulfoxide; Eph, erythropoietin-producing human hepatocellular carcinoma receptor; FRET, fluorescence-resonance energy transfer; GCK, germinal center kinase; Lck, lymphocyte-specific kinase; MD, molecular dynamics; MLK, mixed-lineage kinase; rmsd, root-mean-square deviation; SAR, structure-activity relationship; THF, tetrahydrofuran.

N-terminal domain necessary for ligand binding, whereas its intracellular domain includes a C-terminal domain and a tyrosine kinase domain. Despite the potential therapeutic importance of EphB4, only four series of small molecule inhibitors are currently known (Figure 1).6-9 In 2007, Miyazaki and co-workers reported the synthesis of 3-[4-amino-3-(3-chloro-4-fluorophenyl)thieno[3,2-c]pyridin-7-yl]-benzenesulfonamide (1), which is a potent EphB4 inhibitor.7 One year later, 2,4-bis-anilinopyrimidine derivatives such as 2 showed also high potency as EphB4 inhibitors, and their cocrystallization with human EphB4 highlighted their dual binding mode (Figure 1).10 The marketed drug dasatinib, with Abl1 and Src as primary targets, also showed a very high affinity to Eph kinases.11 High throughput docking is a computational tool frequently used to discover small-molecule inhibitors of enzymes or receptors of known three-dimensional structures.12,13 Recently, we have developed an efficient computational method (termed ALTA for anchor-based library tailoring) to focus a chemical library by docking and prioritizing molecular fragments according to their binding energy.6 From a collection of about 700 000 compounds, ALTA generated a focused library of 21 418 molecules, each containing at least one fragment predicted to bind to the ATP-binding site of EphB4. Automatic docking of these 21 418 molecules yielded two series of micromolar inhibitors, one of them based on a xanthine scaffold predicted to be involved in two hydrogen bonds with the hinge region that connects the N-terminal and C-terminal lobes of the kinase domain. Further characterization of the commercially available 3 (Figure 1) indicated that this molecule binds to the ATP-binding site, as predicted by the docking calculations.6 In addition, an analogue with a pendent anisidine chain (4, Figure 1) showed similar inhibition properties to 3 and was active in a cell-based assay. Here, we present a medicinal chemistry campaign aimed at improving the affinity of the micromolar hits 3 and 4 identified by in silico screening. The optimization was carried out using

r 2009 American Chemical Society

Published on Web 09/29/2009

1. Introduction Angiogenesis, the formation of new blood vessels from preexisting ones, has been identified as one of the key steps in human carcinogenesis. In fact, nutrient supply and waste elimination are required for cell proliferation. Because of low toxicity and resistance potential,1 as well as the possibility of treating a large spectrum of solid tumor types,2 angiogenesis inhibition is considered a promising target in anticancer therapies. Several studies have implicated erythropoietin-producing human hepatocellular carcinoma receptor (Epha) signaling in sprouting angiogenesis and blood vessel remodeling during vascular development.3 Furthermore, overexpression of several of the 14 Eph receptors has been linked to tumors and the associated vasculature, suggesting a critical role in tumorrelated angiogenesis. In fact, inhibition of binding of EphB4 to its natural ligand EphrinB2 using soluble extracellular domains of EphB4 has been shown to reduce tumor growth in murine tumor xenograft models.4,5 Thus, inhibition of Eph angiogenic activity has been recognized as an effective strategy for blocking tumor progression and metastasis. Like all receptor tyrosine kinases, EphB4 is a type-I transmembrane protein. Its extracellular domain is composed of an

pubs.acs.org/jmc

6434 Journal of Medicinal Chemistry, 2009, Vol. 52, No. 20

Lafleur et al.

the binding mode obtained by automatic docking into EphB4. Chemical synthesis of about 30 derivatives of 3 and 4 yielded six nanomolar inhibitors, one of which (compound 66) has an IC50 of 2-5 nM and shows good selectivity against other protein kinases. The addition of only two heavy atoms (-CH3 and -OH substituents at the phenyl ring, 4 vs 66) has resulted in a ∼1000 times improvement of affinity, which is a remarkable example of the usefulness of structure-based hit modifications.

Figure 1. Previously known EphB4 inhibitors.6-9

2. Synthesis Our foreseen modifications to the scaffold of compound 4 (R1-7) for the structure-activity relationship study (SAR) have been summarized at the top of Table 1. Such studies demand a rapid, reliable, and flexible access to a wide variety of potentially active structures with a minimum synthetic variation cost. To succeed in such a goal, a modular synthetic approach was developed. Thus, methyl and benzyl substituted derivatives at R1 were prepared in parallel as summarized in Scheme 1. The synthesis started by condensation of cyanoacetic acid with commercially available methylurea or benzylurea to give the corresponding cyanoacetylurea intermediates, which upon treatment with base afforded the desired 1-alkyl6-aminouracils (5,6).14 Nitrosation at C5 of the pyrimidine ring with sodium nitrite in acetic acid furnished compounds 7 and 8, which were subsequently reduced with sodium dithionite to give 1-alkyl-5,6-diaminouracils 9 and 10.15 The diamino compounds were immediately refluxed with formic acid to give an amide intermediate, followed by cyclization in basic media, to afford the desired xanthines 11 and 12. Bromination at C8 of the xanthine core with Br2 and sodium acetate in acetic acid led to the formation of the key 8-bromoxanthines 13 and 14 in excellent overall yield.16 The synthesis of the noncommercially available R-halo ketones is summarized in Scheme 2. First, commercially available 1,3-benzodioxole-5-carboxaldehyde (piperonal, 15)

Table 1. EphB4 Inhibition Data for Xanthine Derivatives

compd

R1

R2

R3

R4

R5

R6

R7

IC50 (nM)a

3 4 45 46 69 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

Me Me Bn Bn H Me Me Me Me Me Me Me Me Me Me Me Me Me Me Me Me Me Me Me Me

H H H H H H H 1,3-dioxol OMe H H Me H H H OH OH H H H H Me Me Me H

H H H H H 1,3-dioxol 1,3-dioxol H H OMe H H Me Me H H H H H H H OH H H H

H H H F F H H H H H OMe H H H Me H H H H OH OH H OH H H

H H H H H H H H H H H H H H H H H OH OH H H H H OH OH

H H H H H H H H H H H H H H H H H H H H H H H H OH

4-hydroxybutyl o-methoxyphenyl o-methoxyphenyl o-methoxyphenyl o-methoxyphenyl o-methoxyphenyl 4-hydroxybutyl o-methoxyphenyl butyl butyl butyl o-methoxyphenyl o-methoxyphenyl butyl butyl o-methoxyphenyl butyl o-methoxyphenyl butyl o-methoxyphenyl butyl o-methoxyphenyl o-methoxyphenyl o-methoxyphenyl o-methoxyphenyl

7000 (5680) 3300 (4350) >10000 >10000 5400 >20000 42% at 10 μM 30% at 10 μM >10000 36% at 10 μM >10000 180 (47) >10000 37% at 10 μM 38% at 10 μM 64% at 10 μM 1600 368 (213) 691 59% at 10 μM 558 1200 236 5 (1.6) 5000

a IC50 values were measured by a FRET based enzymatic assay while values in parentheses are IC50 values determined by an enzymatic assay with radioactive ATP (see Experimental Section).

Article

was reacted with methylmagnesium bromide to give secondary alcohol 16, which was subsequently oxidized in the presence of manganese oxide to give methyl ketone 17. R-Bromination at the methyl position took place in the presence of phenyltrimethylammonium tribromide to give alkylating agent 18 in only three steps. 2-Bromo-20 -methyl-30 -hydroxyacetophenone (24) and 2-bromo-20 -methyl-40 -hydroxyacetophenone (25) were prepared by treatment of the corresponding carboxylic acids with methyllithium to afford methyl ketones 21 and 23, Scheme 1a

Journal of Medicinal Chemistry, 2009, Vol. 52, No. 20

6435

respectively (Scheme 2). Demethylation of 21 using AlCl3 yielded phenolic derivative 22. Finally, R-bromination at the methyl group was achieved in the presence of copper(II) bromide in chloroform to give monohalogenated ketones 24 and 25. Alkylation at the N7 position of the xanthine core with the appropriate R-haloketone was accomplished using N,Ndiisopropylethylamine in dimethylformamide, providing the 8-bromo-3-alkyl-7-(2-oxo-2-phenylethyl)xanthine derivatives 26-44 in good yields. Treatment of these intermediates with primary alkylamines or aromatic amines in a sealed tube at 180 °C in ethanol as solvent afforded the desired imidazo[1,2-f ]xanthine derivatives 45-68 (Scheme 3). Two more derivatives were prepared as chemical probes to address the binding mode of these molecules to EphB4 (Scheme 4). First, the benzyl group in 46 was removed in the presence of Pd/C with ammonium formate to give 69. Alternatively, the nitrile group in 68 was transformed into the corresponding carboxylic acid (70) by hydrolysis with sulfuric acid. 3. Results and Discussion

a Reagents and conditions: (a) (i) NCCH2CO2H, Ac2O, 60 °C, 1.5 h; (ii) NaOH, 90 °C, 30 min; (b) NaNO2, AcOH-H2O, 25 °C, 12 h; (c) Na2S2O4, NH4OH, 50 °C, 1 h, then 25 °C, 8 h; (d) (i) formic acid, reflux, 3 h; (ii) NaOH, reflux, 1 h; (e) Br2, NaOAc, AcOH, 65 °C, 2 h.

Scheme 2a

a Reagents and conditions: (a) MeMgBr, THF, -10 °C, 1 h; (b) MnO2, Et2O, 25 °C, 48 h; (c) phenyltrimethylammonium tribromide, THF, 25 °C, 16 h; (d) MeLi, Et2O, 25 °C, 3.5 h; (e) AlCl3, PhCl, reflux, 6 h; (f) CuBr2, CHCl3, EtOAc, reflux, 15 h.

The inhibitory activity of the compounds prepared in Schemes 3 and 4 was measured by a fluorescence resonance energy transfer (FRET) based enzymatic assay that quantifies inhibition of phosphorylation of a synthetic substrate of EphB4 at Km concentration of ATP (see Experimental Section). Experimental Validation of the Binding Mode. The binding mode of compound 3 obtained by automatic docking6 Scheme 4a

a Reagents and conditions: (a) Pd/C, ammonium formate, MeOH, 140 °C, 3 h; (b) H2SO4, H2O, 120 °C, 2 h.

Scheme 3a

a

Reagents and conditions: (a) R-halo ketone, DIPEA, DMF, 25 °C, 17 h; (b) primary amine, EtOH, sealed tube, reflux, 15 h.

6436 Journal of Medicinal Chemistry, 2009, Vol. 52, No. 20

Lafleur et al.

Figure 2. Binding modes of compound 3. Poses A (left) and B (right) were generated by automatic6 and manual docking, respectively. In the schematic view (bottom), the side chain of Phe695 is not shown for clarity. Table 2. EphB4 Inhibition Data for Compounds Synthesized To Discriminate between Two Putative Binding Modes

compd

R4

IC50 (nM)

47 68 70

NO2 CN CO2H

>20000 >20000 >20000

indicated that the carbonyl C6dO and amide group N1-H are involved in hydrogen bonds with the backbone polar groups of Met696 (Figure 2, left). In this model, the phenyl ring of 3 is buried in the hydrophobic pocket, while the 4-hydroxybutyl lateral chain at R7 points toward the solvent. Visual inspection of such binding mode suggested that another pose could be obtained by a 180° rotation of the first one (binding mode B, Figure 2, right) so that the phenyl ring points toward the solvent. In pose B, the C6dO and N1-H of compound 3 interact with the NH of Met696 and the backbone CO of Glu694, respectively. To discriminate between these two binding modes, we decided to introduce polar

substituents at the para position of the phenyl ring. Derivatives with a nitro (47), a cyano (68), or a highly hydrophilic carboxylic group (70) were prepared according to the method described in Schemes 3 and 4. These three derivatives were inactive (Table 2). Furthermore, a hydroxyl substituent at R4 reduced the affinity by a factor of about 2-3 (compare 4 and 62, Table 1). These results indicate that the phenyl ring of compound 3 more likely fits into the hydrophobic pocket of the ATP-binding site, as suggested by automatic docking (binding mode A in Figure 2). Lead Optimization Strategy. Once sufficient evidence about the binding mode had been gathered, we started our optimization campaign from commercially available 4 (compound 5 in ref 6). Compounds 3 and 4 differ only at R7 (alkyl vs aromatic) and have similar IC50 values in the enzymatic assay (Table 1), but only the latter showed activity in a cellbased assay (Table 1 in ref 6). According to the binding mode, the substituent at N3 of the pyrimidine ring could be involved in additional interactions with Ala700 (Figure 2) so that we decided to start the chemical edition of the molecule at R1. The methyl group was replaced by a benzyl substituent, causing a major loss in the inhibitory activity of these molecules (45 and 46 vs 4). Compound 69, with a N3-H bond, was not more potent than 4. Since no improvement was achieved by modification of the pyrimidine ring, we decided to focus our efforts on a different region of the molecule.

Article

The ATP-binding site of protein kinases can be divided into five different subpockets.17 The size of the hydrophobic pocket is controlled by a so-called gatekeeper residue, and it is well-known that not only affinity but also selectivity can be improved by fully exploring this site.17 In fact, only around 20% of the 518 human kinases have a small gatekeeper residue. Since EphB4 belongs to this class having a threonine gatekeeper (Thr693), we envisioned that a straightforward strategy to improve affinity would stem from modifications of the substitution pattern at the phenyl ring. In sharp contrast to compound 2 and other previously developed inhibitors of EphB4,8,10 a dioxole ring in relative positions R3, R4 (48 and 49) or R2, R3 (50) of the aromatic ring significantly reduced the activity. A methoxy group with different substitution patterns (51, 52, 53) also suppressed the inhibitory activity of the corresponding molecules. According to binding mode A (Figure 2), the space around the phenyl ring is rather limited, thus restricting the size of the substituents that might improve steric complementarity. Furthermore, the phenyl ring is close to the carbonyl group of Glu664, the hydroxyl and carbonyl groups of Ser757, and the NH group of Asp758, which can act as hydrogen bonding partners. On the basis of these observations, we decided to examine less sterically demanding substituents such as methyl or hydroxyl groups, which are expected to fulfill the nearby hydrogen bonding capacity. Notably, a methyl substituent at position R2 of the benzene ring (54) showed an IC50 value close to 100 nM. In contrast, R3 and R4 methyl substituted derivatives (55-57) were almost inactive. Furthermore, introduction of a hydroxyl group at R5 (60) was also beneficial with an IC50 of about 200-400 nM. Some inhibition activity was also detected for compounds bearing the OH group at R2 (58) and R4 (62) of the phenyl ring. In these cases, replacing the anisidine lateral chain at R7 (58, 60, and 62) for an alkyl one such as butyl (59, 61, and 63) seemed to have only a limited influence in the inhibitory activity of these molecules as previously observed for the commercially available compounds with a propyl or a butyl chain at R7.6 With these results in hand, we decided to explore how the combination of the most favored substitution patterns could influence the inhibitory activity. Thus, the methyl substituent was kept at R2 and a hydroxyl group was added at the relative positions R3 (64), R4 (65), and R5 (66) of the benzene ring. Strikingly, a combination of a methyl and a hydroxyl group at R2 and R5, respectively, yielded compound 66, which has a ∼1000-fold higher affinity compared to the original hits obtained by docking, i.e., compounds 3 and 4. As mentioned above, the addition of only one heavy atom, CH3 at R2, resulted in a factor of 20-100 improvement (54 vs 4, or 66 vs 60). This observation led us to further investigate the role of the methyl group, in particular if it stabilizes the orientation of the phenyl ring required for binding. Conformational analysis was performed on 4, 54, 60, and 66 by exhaustive sampling of the dihedral angles involved in the rotation of the phenyl and o-methoxyphenyl rings (γ1 and γ2 in Figure S1 of the Supporting Information), followed by geometry optimization of the resulting structures using quantum mechanics. For each of the four inhibitors, the local minima are distributed into four sets of conformers, which are separated by rotation barriers. The docked conformation lies in one of these four basins whose local minima have similar energy values (maximal difference in energy of 0.082 kcal/mol for 4, 0.219 kcal/mol for 54, 0.257 kcal/mol

Journal of Medicinal Chemistry, 2009, Vol. 52, No. 20

6437

Table 3. Local Selectivity of Compound 66a kinases

IC50 (nM)

EphA1 EphA2 EphA3 EphA4 EphA5 EphA7 EphA8 EphB1 EphB2 EphB3 EphB4

2.9 2.3 40 3.3 3.0 1118 4.5 1.1 1.2 15 1.6

a

These IC50 values were measured at Reaction Biology Corporation.

for 60, and 0.143 kcal/mol for 66), which suggests a quasiequal distribution of the population in each of the four conformers. Notably, the presence of a methyl group at R2 seems to restrict the accessible conformations more than a hydroxyl group at R5 (compare the plots obtained for 54 and 66 and for 4 and 60 in Figure S1). In addition, the conformational strain, which is the energy difference between the minimized bound conformation and the lowest energy conformation of the isolated ligand, was evaluated for each molecule. The similar values obtained for the strain energy of 54 (0.5 kcal/mol) and 4 (0 kcal/mol), as well as 66 (0.7 kcal/ mol) and 60 (0.3 kcal/mol), indicate that the methyl group contributes to a gain in intermolecular van der Waals energy rather than in strain energy (see also the subsection Binding Mode of Compound 66 Investigated by MD Simulations). During the preparation of this manuscript we discovered in the literature a series of inhibitors of the tyrosine kinase Lck with a 2,4-dianilinopyrimidine scaffold which have a very similar SAR for the phenyl substituents to the one observed for compounds 4, 54, 60, and 66.18 Note that the phenyl substituent in both series of compounds is located in the hydrophobic pocket, but it is connected to the 2-anilinopyrimidine core by a -NH- linker in the Lck inhibitors whereas the phenyl ring is attached directly to the xanthine scaffold in the EphB4 inhibitors described here. Selectivity Profile. To assess the specificity of kinase inhibitors, it is useful to distinguish between local and global selectivity profiles, which reflect the inhibitory activity of the tested compound on a single branch of the kinome dendrogram and on the whole kinome, respectively. The local selectivity of compound 66 was tested against a panel of 11 Eph receptor kinases by an enzymatic assay with [γ-33P]ATP (Reaction Biology Corporation). The IC50 values measured for this compound against 10 of the 11 Ephs are in the low nanomolar range (1-40 nM), while an IC50 value of 1.1 μM is observed for EphA7 (Table 3). These IC50 values are consistent with the very high sequence identity (60-90%) of Ephs and the bulkier gatekeeper residue in EphA7 (isoleucine) with respect to the other Eph kinases (threonine). To evaluate the global selectivity, enzymatic assays (at single concentration of inhibitor) were performed for compounds 66 and 54 using a panel of 85 kinases (National Centre for Protein Kinase Profiling at the University of Dundee, Figure 3). Out of these 85 kinases, only five (EphA2, EphB3, Src, Lck, and Yes1) and three (EphA2, Lck, Yes1) are very strongly inhibited by compounds 66 and 54, respectively (less than 10% activity remaining compared to a DMSO control at 1 μM of 66 and 3 μM of 54; see Supporting Information). It is important to note that these

6438 Journal of Medicinal Chemistry, 2009, Vol. 52, No. 20

Lafleur et al.

Figure 4. Time series of the CR root-mean-square deviation (rmsd) of residues 613-883 of the kinase domain of EphB4 with respect to the X-ray structure (PDB code 2VWX). The loop 772-778 was not taken into account because it is not present in the crystal structure.

Figure 3. Selectivity profile of compound 66. The circles correspond to all kinases tested. Inhibition of activity was measured in enzymatic assays with radiolabeled ATP at Reaction Biology Corporation and University of Dundee for 11 and 85 kinases, respectively, while binding affinity of 50 kinases was measured at Ambit Biosciences Corporation. Enzymatic assays: high, medium, low, and no compound affinity for kinase activity (with respect to DMSO control) of 60%, respectively. Binding assay: high, medium, low, and no compound affinity for kinase activity (with respect to DMSO control) of 30%, respectively. Kinome diagram is reproduced courtesy of Cell Signaling Technology, Inc. (www. cellsignal.com).44

five kinases have a threonine as gatekeeper, as well as CSK, BTK, and HER-4, which show relatively strong inhibition by 66 (between 10% and 50% activity remaining compared to a DMSO control). The local and global selectivity profiles prompted us to test the binding of compound 66 against a set of 49 kinases (Ambit Biosciences Corporation) selected among the nearly 100 that are predicted to have a small gatekeeper based on sequence and structure analysis.19 Interestingly, in competition binding assays, compound 66 shows significant affinity (kinase activity